The Bottom Line: The EU is implementing its AI Act with a focus on systemic risks such as cyberattacks through AI or its uncontrolled deployment.
As of Sunday, the European Commission has put in place regulatory powers to oversee frontier AI models such as OpenAI or Anthropic. A 37-person team in Brussels can now impose fines of up to €15 million or 3 percent of global annual revenue and remove models from the market.
The AI Office of the European Commission, established two years ago, receives concrete enforcement powers as of this week. Unit A3 within the office specifically focuses on monitoring how leading AI companies manage systemic risks. The European AI Act, which entered into force in August 2024, requires providers of frontier models such as OpenAI, Anthropic, and Google to assess and mitigate potential systemic risks.
The Commission has defined four concrete risk categories: AI models used for chemical, biological, or nuclear attacks; loss of control over a model; cyberattacks through AI; and large-scale manipulation by AI. The regulatory tools are tiered: when serious concerns arise, mitigation measures can be requested; in case of non-compliance, fines of up to €15 million or 3 percent of a company’s global annual revenue threaten. Additionally, the Commission can compel companies to restrict, revise, or remove their models from the market.
Since its creation, the A3 team has prepared implementation of these powers. Last summer, the unit published a Code of Practice to facilitate industry compliance – however, Meta refused to sign it. Staffing also presented challenges: the team operated without leadership between mid-2024 and end-2025, until Matthieu Delescluse was appointed head in January. Delescluse is an experienced Commission official with a background in telecommunications policy, roaming charges, and EU funding programmes, but according to his own profile lacks specific AI security experience.
The timing of the new supervisory powers coincides with a period of heightened attention to AI security: recent incidents have occurred in which AI agents spiraled out of control and attacked other systems. This increases pressure on the relatively small Brussels team to set standards for how governments can ensure AI security. With this, the Commission gains the ability to critically examine companies before new models reach the market.
Source: www.politico.eu · Published 29 July 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.